APPENDIX G: LAB SETUP GUIDE

THE RECOMMENDED LAB REPOSITORY LAYOUT This is the folder structure to build up over Chapters 65 through 69. Do not create everything at once, and add each file when its chapter reaches it. resume_review_agent/ main_loop.py: the core think, act, and observe loop, built in Chapter 65 tool_registry.py: the tools read_file, search_job_posting, and write_report, built in Chapter 65 permission_gate.py: restricts file access to the resumes and reports folders, built in Chapter 65 context_builder.py: selects and formats the relevant messages, built in Chapter 66 compaction.py: summarizes history once a token threshold is passed, built in Chapter 66 progress_notes.md: the scratchpad the agent maintains, created in Chapter 66 hooks.py: pre-execution, post-execution, validation, approval, and continuation hooks, built in Chapter 67 mcp_client_config.json: the connection to a local Model Context Protocol server, built in Chapter 67 tracing.py: span and trace instrumentation, built in Chapter 68 eval_set.json: the fifteen item evaluation dataset, built in Chapter 68 dashboard.py: a simple summary of success rate, cost, and latency, built in Chapter 68 supervisor.py, matcher_agent.py, and writer_agent.py: the multi-agent refactor, built in Chapter 69 model_adapter.py: the single doorway to the model, holding the scripted mock model, built in Chapter 65 and replaced in Chapter 69 hosted_gemini.py: optional code that connects the doorway to Google Gemini, built in Chapter 65, Step 3B make_sample_data.py, make_more_data.py, make_injection_data.py, make_empty_data.py, and make_eval_set.py: scripts that write the fictional data and the evaluation set, built in Chapters 65 through 68 test_gate.py: the short test that tries to break the permission gate, built in Chapter 65 mini_mcp_server.py, mcp_client.py, and mcp_demo.py: the teaching tool server, its client, and a demonstration, built in Chapter 67 routing.py: model tiers and the cost ledger, built in Chapter 69 eval_runner.py: runs the evaluation set and scores it, built in Chapter 68 and replaced in Chapter 69 sample_data/ resume_01.txt through resume_25.txt: fictional resumes, built in Chapters 65 and 66 job_posting.txt: the fictional job posting, built in Chapter 65 resume_injection_test.txt: the planted prompt injection sample, built in Chapter 67 resume_empty.txt: the planted runaway loop trigger, built in Chapter 68 reports/: written by the agent at run time logs/: structured run logs and traces, written at run time